Microsoft has introduced MAI-Cyber-1-Flash, a compact in-house security model built for vulnerability discovery and remediation inside MDASH, its multi-agent security harness.

What changed

The company says MAI-Cyber-1-Flash is designed for challenging vulnerability work in complex codebases. In Microsoft's framing, the model is not a standalone chatbot product. It is part of MDASH, where multiple models and more than 100 specialized agents are used to find, validate, and remediate software flaws.

Microsoft also said MDASH with MAI-Cyber-1-Flash and GPT-5.4 reached 96% on its CyberGym evaluation, a vendor-reported benchmark result that should be treated as directional until independently tested. The company says the new mix can handle many security tasks more cheaply than its prior MDASH setup, reserving larger models for harder cases.

Why it matters

The broader launch is Project Perception, an agentic security system that Microsoft says coordinates red team, blue team, and green team agents. The goal is to continuously identify attack paths, triage threats, and turn security findings into fixes while keeping human defenders in control.

Project Perception is scheduled to enter public preview on August 3. Microsoft says it will be integrated across Microsoft Security products and will use a multi-model architecture, including specialized cyber models such as MAI-Cyber-1-Flash.

The practical signal is that Microsoft is moving security AI from assistant-style copilots toward persistent agent teams tied to enterprise telemetry, remediation workflows, and model-routing economics.